Evaluation fail! about stn-ocr

[Bartzi]

Sorry that it took me so long to answer!

That is definitely difficult to say.
What version of MXNet are you using? Did you make any changes in the code?

[ThaiLe189]

Yeah i using another version of mxnet because when i compile mxnet with tag v0.9.3, i get error

The command i was ran:
git clone --recursive https://github.com/apache/incubator-mxnet.git mxnet1
git checkout v0.9.3
#edit config file in mxnet

git submodule update --init --recursive
make -j4 \USE_OPENCV=1 \USE_CUDA=1 \USE_CUDA_PATH=/usr/local/cuda \USE_CUDNN=1 \USE_BLAS=openblas

in image is mxnet1 because i don't want to remove version of mxnet i was compiled

[ThaiLe189]

I realize that compile mxnet with cuda give me that error, i have try compile mxnet with command:
make -j4 \USE_OPENCV=1 \USE_BLAS=openblas
and it work but i can't use my gpu for train
I using Cuda version 10.1 and cudnn 7.6.5

Update i try use Cuda 9.0 but have same error

[Bartzi]

Hmm, I think I can not help with that then, since I only tested the code with the version of MXNet showed in the README.
I suggest that you have a look at my other implementation with Chainer from the paper SEE. This should work better on newer systems, since chainer is a nicer framework.

[Bartzi]

Oh and regarding your last screenshot: What is the actual error? I can only see warnings and that there is an error, while compiling cuddn_convolution

[ThaiLe189]

I compile mxnet v0.9.3 with command: make -j4 \USE_OPENCV=1 \USE_CUDA=1 \USE_CUDA_PATH=/usr/local/cuda \USE_CUDNN=1 \USE_BLAS=openblas
and i get that error, i don't know actual error
So let me check Chainer first
Thank you very much.